Spatially Distorted Signaling: How Opinions Against Wind Infrastructure Delay Our Transition to Renewable Energy

This study analyzes the activity of U.S. wind power plants (2000-2016) to understand how population density, income, and anti-wind opinions impacted wind plant operations. The motivation for this project was to investigate these attributes as they related to local resistance against renewable energy developments, in hopes to understand potential delays in our transition to a more sustainable future.
Statistical Analysis
Logit
Wind Power
Energy Access
Sustainable Energy Decarbonization Transition
Author

Sofia Ingersoll

Published

December 14, 2023

Spatially Distorted Signaling: How Opinions Against Wind Infrastructure Delay Our Transition to Renewable Energy

🗃️GitHub Repo: https://github.com/saingersoll/Spatially-Distorted-Signaling-US-Wind-Infrastructure

Overview

Spatial Distorted Signalling (SDS) describes the mobilization of minority opinion holders to electorally push-back and promote legislation that aligns with their beliefs. Leah Stokes (et.al) has explored the SDS phenomenon as a natural experiment in her piece, Electoral Backlash against Climate Policy: A Natural Experiment on Retrospective Voting and Local Resistance to Public Policy (2016). The findings in this paper describe that rural Canadian communities had a greater ability to mobilize and organize political push back against majority chair holders in parliament after the passing of legislation which invited the development of wind infrastructure through incentives.

Stokes’ subsequent work, Replication Data for: Prevalence and Predictors of Wind Energy Opposition in North America (2023), further explores variables like political affiliation and local mobilization. This study aims to reproduce these outcomes with U.S. wind plant data to understand how population density, income, and anti-wind opinions impact wind plant activity, providing insights into local resistance and its effects on renewable energy projects.

Main Takeaways

  • Population Density:

    A unit increase in population density is associated with a slight increase in the odds of having an operational wind plant. This suggests that areas with higher population densities are marginally more likely to host wind plants and less likely to experience minority holder opinions taking the majority. Therein, as urbanization and population density increase, it becomes increasingly feasible and advantageous for these areas to invest in and support wind energy projects.

  • Median Income:

    Higher income bracket wealthier areas may exert more influence on energy policies due to their resources, potentially impacting local energy investments. Higher-income areas may use their financial resources to affect energy policy decisions, potentially hindering wind infrastructure development. Areas with higher incomes might prioritize different types of energy investments based on their resources and political connections. Lower income bracket individuals in lower income brackets may have less influence and lower advocacy for wind infrastructure.

    An increase in median income is linked to a decrease in the odds of having an operational wind plant. Higher income areas show a lower likelihood of wind plant activity, potentially due to different local priorities or economic factors. Uneven socio-economic power-dynamics could lead to minority opinion holders preventing the development of wind power infrastructure, alongside other renewable energy solutions. Overall, this suggests that socioeconomic factors play a influential role in shaping the types of renewable energy projects that are pursued and highlights the need for tailored strategies to address diverse community priorities in renewable energy planning.

  • Anti-Wind Infrastructure Opinion:

    Areas with higher opposition to wind infrastructure are less likely to have operational wind plants. This aligns with expectations that local opposition impacts the establishment of wind plants. Therefore, addressing and mitigating local resistance is essential for enhancing the feasibility and acceptance of wind energy initiatives.

The world’s largest wind turbine, MySE-16-260. Includes a rotor diameter of 853 feet to produce 16-megawatt, located in Pingtan, Fujian Province, China (Yang, Getty Images).

U.S. Wind Power Plant Data

The data source that was utilized in this project, US Wind Data, focuses on the public stance on wind infrastructure for census tract regions within a 3 km buffer zone of a wind infrastructure project. It contains categorical variables, binary variables, continuous socioeconomic factors such as % of races, % precinct political GOP affiliated voting share, mobilization tactics, and more. For simplicity’s sake, we’re going to focus on the variables below.

Variables of Interest:

Name Description
status Describes the project operating status. In this study, we have converted it into a binary variable: 1 is operating, 0 is not_operating.
pop_den Tract-level 2010 census data for population density (per mi^2)
med_inc Tract-level 2010 census data for median income ($)
is_anti_wind Binary measure of wind opposition: 1 is against wind power developments, 0 is pro wind power developments.

U.S. Wind Power Plant Locations

Mapping Wind Power Infrastructure Plants in the U.S.

Before diving in, let’s get a sense of where we’ll be investigating. Using `ggplot()`, we can visualize the locations of wind infrastructure power plants throughout the United States. To achieve a more granular map, we’ll need to utilize another data set to create a base layer for our map in order to observe these wind plants with respect to state and county jurisdictions.

Below we will use the package sf to convert the lat/long vector data into a raster geometry column. In this single line, we will also be assigning the CRS EPSG:4326 to the sf data frame. Coordinate Reference Systems, CRS, are required in order for the data to be projected onto a map. The CRS was selected because it provides a relatively proportionate display of the United States. We are open to suggestions regarding our CRS if a different project better fits our data.

Code 
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----       Load Libraries      ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# The following libraries were selected based on their functionality and ability to optimize our data for mapping.

# Loading Libraries
library(tidyverse)        # essential r package 
library(sf)               # package simplifies spatial dataframes
library(tmap)
library(terra)
library(broom)
library(stars)
library(sjPlot)
library(naniar)
library(cowplot)
library(leaflet)
library(maptiles)       
library(ggthemes)
library(ggspatial)
library(patchwork)
library(kableExtra)

set.seed(99)
knitr::opts_chunk$set(echo = T, warning = F, message = F)

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----        Read & Raster      ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# reading in & storing data
wind_data <- read.csv("../data/wind_data/wind_data_usa.csv")  

# Confirm the Data Loaded Properly
#head(wind_data)                  # displays the first 6 rows of the data

# Let's read in our data
wind_sf <- wind_data %>%       
  # creates geometry column with desired crs 
  st_as_sf(coords = c("longitude", "latitude"), crs = 4326) 
                                     
# quick CRS check
#glimpse(crs(wind_sf))                  # output should reveal WGS84, EPSG:4326

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----        Check Point!       ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Let's stop and see if our outputs are what we expect.
# Were the lat/long columns correctly converted into a geometry column?
# setdiff() is a way to quickly determine the differences between two data sets.

# Sweet! we are looking good
#setdiff(colnames(wind_sf), colnames(wind_data))

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----        Map Wind Plants    ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# First visual of the U.S. wind data provided by the geometry points
wind_plants <- ggplot(wind_sf) +
  annotation_map_tile(type = "osm") +
  geom_sf(col = 'darkgreen',
          alpha = 0.5,
          size = 3)

wind_plants

Visualizing the Data Distribution

Before jumping into any analysis, it’s important to get a sense of how the data is distributed and if there are any underlying trends or biases. The two visual aids we’re going to create are a violin plot with jitter points (left) and a comparative regression plot using OLS and GLM (right). Combining two figures provides us fuller insights into both the general trend and changes in probability of the binary outcome for the population density predictor.

We will employ a series of models to describe the effect of census tract level population density on the operating status of wind power infrastructure. A combination of binary and interaction logit regression will be considered. The initial model will apply OLS regression, this is really a formality to demonstrate why OLS is not the correct approach for interpreting our relationships of interest. The following will be a model with two continuous variables. Binary Indicator Variable will be status column: opertating is 1, and not_operating will be 0.

These variables focus more on regionally dependent factors that intuitively seem to have an impact on mobilization variables that we don’t have time to cover in this project. We’ll be working with a mix of discrete and continuous data, so there some wrangling will be necessary to run the regressions we’re interested in.

Code 
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ---- Inspect & Standarize Data ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Determining Variable Assignments for OLS
#unique(wind_sf$status)     # displays unique values in this

# Need to rename status output variables
# creating two categories: operating & not_operating
# We are removing 'Operating | Decommissioned' because it skews the data
unwanted_status <- "Operating | Decommissioned"
replacement_status <- "Uncertain Status"
wind_sf$status[wind_sf$status== unwanted_status]<-"Uncertain Status"  

# were we successful ?
#unique(wind_sf$status)     # displays unique values in this

# cleaning out NAs for OLS
wind_sf <- wind_sf %>%
  filter(is.na(status) == 'FALSE') %>% 
  filter(is.na(is_anti_wind) == 'FALSE') %>% 
  filter(is.na(pop_den) == 'FALSE') %>% 
  filter(is.na(med_inc) == 'FALSE') %>% 
  filter(is.na(median_age) == 'FALSE') %>% 
  filter(is.na(n_turbs) == 'FALSE')

# were we successful ?
#unique(wind_sf$status)     # displays unique values in this

# if_else preserves the data type but replaces unwanted values
wind_us <- wind_sf %>% 
  mutate(status = if_else(
    status %in% c('Cancelled', 'Out of service (temporarily)', 'Standby', 'Decommissioned', 'Uncertain Status'), 'not_operating',
    'operating') 
  )

# are our only outputs "operating" and "not_operating"?
#print(unique(wind_us$status))

# status as factor and reassigned values
wind_us <- wind_us %>% 
  mutate(status = case_when(status == "operating" ~ 1,
            status == "not_operating" ~ 0))

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----        Check point!       ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# are our only outputs 0 or 1?
paste("The indicator column contains", {unique(wind_us$status)})
[1] "The indicator column contains 1" "The indicator column contains 0"

The visualizations display the majority of the distribution lies within the actively operating wind infrastructure plants. This is an early indication of the degree of sample bias influencing our analysis. A trend of inactive plants and lower population density is notable in both figures. Collectively they demonstrate smaller population densities contain more inactive wind infrastructure plants. This could be attributed to with weight of a singular vote in regions with smaller demographics.

Local mobilization of minority opinion holders in these regions have a greater availability to push back against policymakers. However, this visual does not encapsulate all of the necessary information required to determine this with full certainty. Our data set has low availability for non-operating infrastructure and as such, in the regression figure on the right these are being treated as outliers.

Code 
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----           Violin Distribution          ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Create the violin plot with log scale
density_plot <- ggplot(data = wind_us, 
                       aes(x = factor(status), 
                           y = pop_den, 
                           fill = factor(status))) + 
  
  geom_violin(alpha = 0.6, color = "darkblue") + 
  
  geom_jitter(col = "#F84C0B",
              width = 0,
              height = 0.05,
              alpha = 0.35,
              size = 4) +
  
  labs(title = "Population Density vs Wind Power Plant Operating Status",
       subtitle= "Logorithimic Distribution",
       x = "Activation Status",
       y = expression("Population Density (Log Scale, " ~ mi^-2 ~ ")")) + 
  
  
  # rename x-axis labels for clarity
  scale_x_discrete(labels = c("0" = "Inactive", "1" = "Active")) +  
  # Apply logarithmic scale to y-axis
  scale_y_log10() + 
  
  theme_538() + 
  
  scale_fill_manual(values = c("skyblue", "darkblue")) +
  
  # Adjust title font and alignment
  theme(plot.title = element_text(size = 40,
                                  family = "Georgia", 
                                  face = "bold",
                                  hjust = .99,
                                  color ="#293F2C"),  
        
        # Adjust subtitle font and alignment
        plot.subtitle = element_text(size = 38,
                                     family = "Georgia",
                                     color ="#293F2C",
                                     hjust = 0.5), 
        
        axis.title = element_text(size = 36,
                                  family = "Georgia",
                                  color ="#293F2C"),
        
        axis.text = element_text(size = 34,
                                 family = "Georgia",
                                  color ="#293F2C"),
        
         # Move legend to the bottom
        legend.position = "top", 
        
        # Remove legend title if not needed
        legend.title = element_blank(),  
        
        # Adjust legend text size
        legend.text = element_text(size = 34,
                                   family = "Georgia",
                                   color ="#293F2C"), 
        
         # Background color for legend
        legend.key = element_rect(fill = "grey94", color = "grey94"), 
        
        plot.background = element_rect(color = "#FDFBF7")
        ) +  
  
  coord_flip()

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----        Jitter OLS + GLM                ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Optimized jitter plot with smooth lines
jitter_plot_optimized <- ggplot(data = wind_us, 
                                aes(x = pop_den, 
                                    y = status)) + 
  
  # assign color to legend 
 geom_jitter(aes(color = "Data Points"),
              width = 0,
              height = 0.05,
              alpha = 0.6,
              size = 4) +  # Adjusted size for better visibility
  
  geom_smooth(method = "lm",
              se = FALSE,
              aes(color = "OLS Line"),
              size = 1.2,  # Slightly thicker line for visibility
              linetype = "solid") + 
  
  geom_smooth(method = "glm",
              se = T,
              aes(color = "GLM Line"),
              size = 1.2,
              linetype = "dashed",
              method.args = list(family = "binomial")) +
  
  labs(title = "Population Density vs Wind Power Plant Operating Status",
      subtitle= "Logorithimic Distribution Regression Comparison",
       y = "Activation Status",
       x = expression("Population Density (Log Scale, " ~ mi^-2 ~ ")")) + 
  
  # rename yaxis labels for clarity
  scale_y_continuous(breaks = c(0, 1),
                     labels = c("0" = "Inactive", "1" = "Active")) + 
  
  scale_x_log10() + 
  
  theme_538() +
  
  # Adjust title font and alignment
  theme(plot.title = element_text(size = 40,
                                  family = "Georgia", 
                                  face = "bold",
                                  hjust = .99,
                                  color ="#293F2C"),  
        
        # Adjust subtitle font and alignment
        plot.subtitle = element_text(size = 38,
                                     family = "Georgia",
                                     hjust = 0.5,
                                     color ="#293F2C"), 
        
        axis.title = element_text(size = 36,
                                  family = "Georgia",
                                  color ="#293F2C"),
        
        axis.text = element_text(size = 34,
                                 family = "Georgia",
                                  color ="#293F2C"),
        
         # Move legend to the bottom
        legend.position = "top", 
        
        # Remove legend title if not needed
        legend.title = element_blank(),  
        
        # Adjust legend text size
        legend.text = element_text(size = 34,
                                   family = "Georgia",
                                  color ="#293F2C"), 
        
         # Background color for legend
        legend.key = element_rect(fill = "grey94", color = "grey94"), 
        
        plot.background = element_rect(color = "#FDFBF7")
        ) +   
  
   scale_color_manual(name = "Legend",  # Title for the legend
                     values = c("Data Points" = "#F84C0B", 
                                "OLS Line" = "blue", 
                                "GLM Line" = "skyblue"),
                     labels = c("Data Points" = "Data Points", 
                                "OLS Line" = "OLS Line", 
                                "GLM Line" = "GLM Line"))


# Combine plots horizontally
combined_plot <- density_plot + 
                  jitter_plot_optimized + 
  # Arrange plots side-by-side
                  plot_layout(ncol = 2)  

combined_plot

Comprehensive Interaction Model Containing Binary Predictor Variable

Coefficient and Odds Ratio Table

To interpret the model, we compute the odds ratios for each coefficient, providing insight into how each variable and its interactions affect wind plant activity.

\[\operatorname{logit}(p)=\log \left(\frac{p}{1-p}\right)=\beta_0+\beta_1 (Population Density) * \beta_2 (Median Income) * \beta_3 (AntiWindOpinion) +\varepsilon \]

This table provides a comprehensive view of how each variable and their interactions contribute to the likelihood of wind plant activity, facilitating a better understanding of the model’s results.

  • Main Effects: The odds ratios for pop_den, med_inc, and is_anti_wind indicate the individual effects of each variable on the likelihood of having an active wind plant.
  • Interaction Effects: The interaction terms show how the relationship between each pair of variables influences the odds of wind plant activity. For instance, the interaction between pop_den and med_inc captures how the effect of population density on wind plant status changes with median income.
Code 
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# ----     Comprehensive Model   ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Fit the comprehensive model
comprehensive_model <- glm(status ~ pop_den * med_inc * is_anti_wind,
                           data = wind_us,
                           family = 'binomial')

# Extract coefficients and compute odds ratios
coef_summary <- data.frame(
  Term = names(coef(comprehensive_model)),
  Estimate = coef(comprehensive_model),
  Odds_Ratio = exp(coef(comprehensive_model))
)

# Create and style the table
kable(coef_summary, format = "html", digits = 4, caption = "Table 1: Summary of Coefficients and Odds Ratios") %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed", "responsive"), position = "center", font_size = 12)
Table 1: Summary of Coefficients and Odds Ratios
Term Estimate Odds_Ratio
(Intercept) (Intercept) 4.1017 60.4441
pop_den pop_den 0.0011 1.0011
med_inc med_inc 0.0000 1.0000
is_anti_wind is_anti_wind -0.7008 0.4962
pop_den:med_inc pop_den:med_inc 0.0000 1.0000
pop_den:is_anti_wind pop_den:is_anti_wind -0.0016 0.9984
med_inc:is_anti_wind med_inc:is_anti_wind 0.0000 1.0000
pop_den:med_inc:is_anti_wind pop_den:med_inc:is_anti_wind 0.0000 1.0000

References

Data Citation

Replication Data for: Prevalence and Predictors of Wind Energy Opposition in North America: DOI 10.7910/DVN/LE2V0R

  • Collaborators: Leah Stokes, Emma Franzblau, Jessica R. Lovering, Chris Miljanich

  • Source: American Wind Association, Columbia Sabin Center

Citations

  1. Stokes, Leah C., et al. “Prevalence and Predictors of Wind Energy Opposition in North America.” Proceedings of the National Academy of Sciences, vol. 120, no. 40, Sept. 2023, doi:10.1073/pnas.2302313120.
  2. Stoke, L. C. “Front Matter.” American Journal of Political Science, vol. 60, no. 4, 2016. JSTOR, http://www.jstor.org/stable/24877456. Accessed 26 October. 2023
  3. Kahn, M. E. (2007). Green Cities: Urban Growth and the Environment. Brookings Institution Press.
  4. McCright, A. M., & Dunlap, R. E. (2011). The Politicization of Climate Change and Polarization in the American Public’s Views of Global Warming. Society & Natural Resources, 24(5), 398-413.
  5. Verba, S., Schlozman, K. L., & Brady, H. E. (1995). Voice and Equality: Civic Voluntarism in American Politics. Harvard University Press.
  6. Hertel, T. W., & Tsigas, M. E. (2002). The Role of Political Lobbying and Financial Contributions in Policy Making. In Public Policy Analysis (pp. 67-90). Routledge.
  7. Kirschen, D. S., & Strbac, G. (2004). Fundamentals of Power System Economics. Wiley.
  8. Gillingham, K., & Stock, J. H. (2018). The Cost of Reducing Greenhouse Gas Emissions. Journal of Economic Perspectives, 32(4), 169-190. doi:10.1257/jep.32.4.169
  9. Lund, H., & Mathiesen, B. V. (2009). Energy system analysis of 100% renewable energy systems: The case of Denmark in years 2030 and 2050. Energy, 34(5), 524-532. doi:10.1016/j.energy.2008.10.009
  10. Krohn, S., & Damborg, S. (1999). Public attitudes towards wind power. Renewable Energy, 16(1), 954-960. doi:10.1016/S0960-1481(98)00339-5
  11. Wolsink, M. (2007). Wind power implementation: The nature of public beliefs and the social acceptance of renewable energy. Renewable and Sustainable Energy Reviews, 11(6), 1586-1603. doi:10.1016/j.rser.2005.10.001
  12. Stokes, L. C., & Breetz, H. L. (2018). Public opinion and wind energy: The impact of socio-economic factors on policy support. Environmental Politics, 27(1), 1-22. doi:10.1080/09644016.2017.1387160
Back to top

Citation

BibTeX citation:
@online{ingersoll2023,
  author = {Ingersoll, Sofia},
  title = {Spatially {Distorted} {Signaling:} {How} {Opinions} {Against}
    {Wind} {Infrastructure} {Delay} {Our} {Transition} to {Renewable}
    {Energy}},
  date = {2023-12-14},
  url = {https://saingersoll.github.io/posts/SDS_Wind_Infrastructure},
  langid = {en}
}
For attribution, please cite this work as:
Ingersoll, Sofia. 2023. “Spatially Distorted Signaling: How Opinions Against Wind Infrastructure Delay Our Transition to Renewable Energy.” December 14, 2023. https://saingersoll.github.io/posts/SDS_Wind_Infrastructure.